Head landmarks quantized

April 16, 2026 ยท View on GitHub

Use case : Pose estimation

Model description

Head landmarks is a single pose estimation model targeted for real-time processing implemented in ONNX.

The model is quantized in int8 format using onnx quantizer.

Network information

Network informationValue
FrameworkONNX Runtime
Quantizationint8
Provenancehttps://github.com/PINTO0309/PINTO_model_zoo/tree/main/032_FaceMesh
Paperhttps://developers.google.com/ml-kit/vision/face-mesh-detection

Networks inputs / outputs

With an image resolution of NxM with K keypoints to detect :

Input ShapeDescription
(1, N, M, 3)Single NxM RGB image with UINT8 values between 0 and 255
Output ShapeDescription
(1, 1, 1, Kx2)FLOAT values Where Kx2 are the (x,y) values of each keypoints
PlatformSupportedRecommended
STM32L0[][]
STM32L4[][]
STM32U5[][]
STM32H7[][]
STM32MP1[x][]
STM32MP2[x][x]
STM32N6[x][x]

Performances

Metrics

Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.

Reference NPU memory footprint

ModelFormatResolutionSeriesInternal RAM (KiB)External RAM (KiB)Weights Flash (KiB)STEdgeAI Core version
head_landmarksInt8224x224x3STM32N61739.50.03246.474.0.0

Reference NPU inference time

ModelFormatResolutionBoardExecution EngineInference time (ms)Inf / secSTEdgeAI Core version
head_landmarksInt8224x224x3STM32N6570-DKNPU/MCU5.48182.44.0.0